skills/orchestra-research/ai-research-skills/distributed-llm-pretraining-torchtitan/Gen Agent Trust Hub
distributed-llm-pretraining-torchtitan
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- EXTERNAL_DOWNLOADS (SAFE): The skill clones the 'torchtitan' repository and installs the 'torchao' library from GitHub. Both sources belong to the 'pytorch' organization, which is a verified and trusted GitHub organization, ensuring the integrity of the downloaded code.\n- COMMAND_EXECUTION (SAFE): Training and setup procedures utilize 'torchrun' and local scripts (e.g., 'run_train.sh', 'download_hf_assets.py') as intended for the skill's primary purpose. These commands are transparent and align with standard distributed training workflows.\n- CREDENTIALS_UNSAFE (SAFE): The skill correctly handles sensitive information, such as Hugging Face tokens, by utilizing clear placeholders like 'YOUR_HF_TOKEN' and instruction-based input rather than hardcoding any secrets.\n- DYNAMIC_EXECUTION (SAFE): The use of 'torch.compile' is a standard performance optimization within the PyTorch ecosystem and does not involve the execution of untrusted or dynamically generated strings.
Audit Metadata